Optical Character Recognition

Even the most advanced imaging techniques will not recover all of the undertexts in the Archimedes Palimpsest, because sometimes these texts are simply no longer present. Individual characters have been partly erased, and sometimes completely obliterated. It is proposed that computers could serve as valuable tools in reconstructing partial characters when ink is missing, or when characters have been obscured by the prayerbook text. In recent years much progress has been made in making computers compare images of faces with those in a database, and finding a match. If this is possible with images of the human face, the thought is that it should be possible to achieve a similar result with the limited range and very particular shapes of the Byzantine script that was employed by the scribe of the Archimedes text. It also significantly helps that only one scribe wrote the Archimedes text, and that he had a very regular hand — that is, his letter shapes are consistent.

The limits of optical character recognition by computer are well understood: the best that a computer can come up with are probabilities that a certain shape is a particular Greek character. These probabilities can be narrowed down both by looking at the spatial characteristics of a character, and then by considering the likelihood that this character would be preceded or would be followed by a particular identified letter. The results will facilitate scholars by presenting them with a range of possibilities from which they might choose.

The optical character recognition has two distinct advantages. Firstly, compared to the advanced imaging techniques that are being employed, it is inexpensive. Secondly, the character recognition is not based upon the manuscript itself, but upon the best available images of the manuscript. Therefore this investigation does not rely on access to a fragile, unique, and ancient document.

The Archimedes project asked for submissions. The guidelines for the submission were very specific: to provide a tool that a scholar could use to give practical help in the decipherment of the Archimedes text.

The submission that the owner chose to fund was provided by Derek Walvoord (left). The interface that he devised is depicted below.

Scholars deemed that the machine had potential for further development, and the owner has agreed to fund the ongoing character recognition effort.